Any thoughts on parallel signals processing?

[OUCyf]

I found dascore used scipy to do signal processing, some notes mentioned that we can use jax or others, so how is it going? Recently I used kfr (https://github.com/kfrlib/kfr) and openmp for signal processing for my ambient noise packages which are also for das processing, and from the preliminary comparison results, kfr is faster than scipy in detrend, filter, taper and ..., the best feelings is that it can use openmp in c++.

I provide a Python interface for those functions via xtensor, and I am trying to publish this ambient noise package in pypi and conda-forge recently, and we can see it in a few days maybe. If it is possible I can also provide this C++ interface for dascore in the future.

BTW i can't use it in win64, but mac and linux is good.

[d-chambers]

Hi @OUCyf,

Sorry for the late reply, for some reason I didn't get a notification about this one.

I think it sounds great to provide a dascore interfaces for your library. Probably the easiest way is for your new package to accept a dascore.Patch object as a potential input, then return patch object. The dascore.patch_function might be useful here.

If you don't want to take dascore as a dependency it could also be used like this:

import dascore
from my_fast_dsp_library import processing_function

patch = dascore.get_example_patch()
processed_patch = patch.new(data=processing_function(patch.data))

The new routines probably can't be the default processing in DASCore, however, since we do need to support windows and scipy, although slower, is tried-and-true.

Thanks for posting the discussion and please let me know when you publish your package, I would love to try it out.